Stereo Matching Method Research Based On Mobile Robot With Binocular Stereo Vision

Since the appearance of mobile robots, they have taken great convenience to thelife and production for people. With the wide use of mobile robots in different fieldstogether with the development of vision technology, people have paid more and moreattention on the research of binocular stereo vision on the mobile robot. Among this,stereo matching method is the key technology in the binocular stereo vision, aeffective matching method has a high demand for the matching accuracy, so imagematching technology is still a problem in the development of binocular vision, thistechnology is being continually researched and improved.Binocular stereo vision technology has been widely applied into home robots now,which provides important help to mobile robot identification and self-localization inthe indoor environment. In order to make mobile robots finish a better assignmentindoor, it demands that mobile robots can realize the accurate localization in acomplex indoor environment, but the layout of reference objects such as table, chair isoften changed by external factors, which is not good for robot accurate localization. Inorder to realize the reliable and accurate localization of mobile robots in the natureenvironment indoor, this paper chose the ceiling corner which is not easily affected bythe environment change as the reference object to research binocular stereo matchingmethod, aimed to improve matching accuracy and real-time and finally providedreliable evidences for mobile robot accurate localization indoors.This paper introduced a series of image processing methods such as imagereprocessing operations including image gray processing method, image filteringmethod and used these methods to process the images for the next operation. It alsoshortly introduced some kinds of matching methods which are often used. Accordingto the research object in this paper, this paper used SIFT operator and Harris operatorto detect feature points in ceiling images, proposed different matching methods forthese two operators respectively. Firstly, this paper researched and analyzed the theoryof SIFT operator, used Kd-tree algorithm and RANSAC algorithm to realize thewhole matching method based on SIFT operator. After that, this paper introducedHarris operator, used NCC algorithm and RANSAN algorithm to realize the wholematching method based on Harris operator. In order to improve the accurate andreal-time matching of the corner in the image, this paper improved the matchingmethod based on Harris operator. This paper gained the image of ceiling corners indifferent positions by using the binocular vision system on Voyager II Robot. Finally, through the comparative experiments among these three matching methods ondifferent ceiling corner images in different positions, it proved that the result broughtimproved method could realize a better accurate matching for the ceiling cornerindoor.